ECMNet: A deep framework for copy move forgery localization detection and classification
Year: 2025
Type: article
Abstract: Copy-move forgery is one of the most prevalent forms of digital image forgery. It violates the authenticity of digital images in the context of documentation, judicial reports, social media, military operations, journalism, intelligence, and so on. However, the localization of a forged region in an image has several key challenges, such as false bounding boxes and inaccurate detection. Moreover, the accuracy of traditional methods relies on handcrafted feature extraction which requires a high level of human involvement. To address these challenges, an enhanced center mask regional convolutional neural network (ECMNet) has been proposed. To advance the efficacy of the proposed ECMNet, a modified backbone integrating MobileNetV2 and a convolutional block attention module has been proposed. Additionally, a Huber loss based a bounding box loss function is proposed to achieve precise localization. The effectiveness of the present work lies in the new combination of these three components in one ECMNet framework. Later, a web application interface for the proposed ECMNet has been developed in Django for real-time forgery detection and localization. The performance has been evaluated for publicly available multi- datasets such as MICC-F220, modified CASIA, CoMoFoD, COVERAGE, and forgery image datasets. The proposed ECMNet shows better accuracy, surpassing the baseline model, Mask-RCNN, Mask Score-RCNN, Center Mask-RCNN and cascade Mask-RCNN by 17.1 %, 12.2 %, 9.6 %, and 7.2 %. The comparative visual and statistical performance reveals that the proposed ECMRNet facilitates the precise localization, detection, and classification copy-move image forgery.
Source: Results in Engineering
Authors: Anjani Kumar Rai, Subodh Srivastava
Institutions: National Institute of Technology Patna, GLA University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $1,940
OpenAlex ID: W4417086089